Research On Vector Quantization Technology Of Low Bit Vocoder Based On MELP

With the development of communication technology, the proportion of non-voice information such as images and data in information transmission is increasing. However, voice communication is still the main function of most communication systems. In the terrestrial cellular mobile communications and satellite mobile communication systems, the contradiction between the high speed growth of voice data and the limited bandwidth resource is becoming more and more serious. The application of high quality and low bit rate speech coder is one of the main way to improve the bandwidth efficiency and increase the capacity of communication system, efficient speech coding algorithm is the research emphasis and hotpots in recent years.As an efficient data compression technology, Vector quantization plays an important role in speech coding field for is high compression ratio and simple decoding. the performance of vector quantization directly affects the quality of synthesized speech. Hence, in this thesis, we deeply study the basic principles of vector quantization technology and research the characteristics of Particle Swarm Optimization(PSO) and Chaos algorithm, then combined their advantages in optimization, a new codebook design algorithm based on chaotic particle swarm optimization is proposed aiming at the deficiencies of the traditional vector quantization codebook design algorithm. Using the proposed algorithm, the speech characteristic parameters codebook is designed. And the codebook of Linear Spectrum Frequency(LSF) has been applied to 2.4kbps Mixed Excited Linear Predictive(MELP) vocoder. Compared to LBG and PSO algorithm, the codebook designed by the proposed algorithm achieves the lowest averaged spectral distortion of synthetic speech, which indicates that the codebook designed through Chaotic Particle Swarm Optimization perform well.In this thesis, making a deep study of the basic principles and coding process of 2.4kbps MELP vocoder.Based on this, we made full use of the correlation between speech frames and adopted multi-frame joint quantization technology to reduce the bit rate of speech vocoder and implement a 1.2kbps low rate vocoder. The proposed codebook design algorithm based on chaos particle swarm optimization is used for designing the vector quantization codebook of residuals harmonic amplitude and linear spectrum frequency.The performance of vecoder synthetic speech are tested through Mean Opinion Score(MOS) and Diagnostic Rhyme Test(DRT), the results of which show that the speech synthesized by the 1.2kbps vocoder has high clarity and intelligibility.It also proves that the improved vector quantization algorithm proposed in this paper has a broader application prospect.